Three Orthogonal Dimensions for Psychoacoustic Sonification
Total Page:16
File Type:pdf, Size:1020Kb
Journal Title XX(X):1–12 Three Orthogonal Dimensions for c The Author(s) 2016 Reprints and permission: Psychoacoustic Sonification sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/ToBeAssigned www.sagepub.com/ SAGE Tim Ziemer12 and Holger Schultheis13 Abstract Objective: Three perceptually orthogonal auditory dimensions for multidimensional and multivariate data sonification are identified and experimentally validated. Background: Psychoacoustic investigations have shown that orthogonal acoustical parameters may interfere perceptually. The literature hardly offers any solutions to this problem, and previous auditory display approaches have failed to implement auditory dimensions that are perceived orthogonally by a user. In this study we demonstrate how a location in three-dimensional space can be sonified unambiguously by the implementation of perceptually orthogonal psychoacoustic attributes in monophonic playback. Method: Perceptually orthogonal auditory attributes are identified from literature research and experience in music and psychoacoustic research. We carried out an experiment with 21 participants who identified sonified locations in two-dimensional space. Results: With just 5 minutes of explanation and exploration, naive users can interpret our multidimensional sonification with high accuracy. Conclusion: We identified a set of perceptually orthogonal auditory dimensions suitable for three- dimensional data sonification. Application: Three-dimensional data sonification promises blind navigation, e.g. for unmanned vehicles, and reliable real-time monitoring of multivariate data, e.g., in the patient care sector. Keywords Auditory Display, Audition, Noise/acoustics, Sound Design, Interpretability Introduction Noirhomme-Fraiture et al.(2008); Bly(1982); Stewart (2010); Rabenhorst et al.(1990); Hermann(2002). Sonification is a powerful means to complement or replace visual displays, especially in situations in which vision is Here, perceptual orthogonality means that while two quan- limited (e.g., in darkness, fog, smoke, muddy waters, etc. or tities are simultaneously sonified, both can be interpreted. due to occlusion), in which the visual scene is overloaded Moreover, if one quantity changes, the change of sound (e.g. due to too many displays or visual distractors), or in can be attributed to its corresponding quantity, and unam- which spatio-visual processing is the bottleneck of spatial biguously interpreted. This obvious necessity is not eas- cognition Walker and Nees(2011). ily achieved. Due to the complicated, nonlinear processing of the auditory system, all physical sound field quantities There is a need for orthogonal dimensions in sonification can affect practically all perceptual attributes of sound. for multidimensional or multivariate data Neuhoff et al. Despite its importance, the lack of perceptual orthogonality (2002); Yeung(1980); Barrass(1997); Watson and is considered one of the most challenging issues in sonic Sanderson(2004); Worrall(2019)(Worrall 2009, ch. 6). interaction design, auditory interfaces for Human-Computer The most prominent application area for multidimensional Interaction, auditory display, and, especially, sonification sonification is spatial navigation, e.g., in real and virtual design Visell et al.(2013); Brewster(2003); Worrall(2009); environments Ziemer and Schultheis(2018b); Lokki and Hermann(2002); Anderson and Sanderson(2009, 2004); Grohn¨ (2005); Walker and Lindsay(2006), games Degara arXiv:1912.00766v1 [cs.SD] 28 Nov 2019 Neuhoff(2011); Kramer(1994); Grond(2013). et al.(2013a), piloting Towers et al.(2014); Florez(1936), remote vehicle control Vasilijevic et al.(2016), autonomous In this paper we present auditory attributes that can serve drivingGray(2011), image-guided surgical interventions as three orthogonal dimensions. The approach is evaluated in Black et al.(2017); Ziemer and Black(2017a); Ziemer et al. a listening test with naive listeners. (2017) and neuronavigation Willems et al.(2005). Degara et al.(2014) even consider sonification for navigation “one of the most important tasks in auditory display research”. Besides navigation, examples for multidimensional or multivariate data sonification include motion analysis and interactive feedback in sports training and neuromotor 1University of Bremen, Bremen Spatial Cognition Center rehabilitation Huang et al.(2006); Scholz et al.(2014); 2Medical Image Computing Group 3Institute for Artificial Intelligence Reh et al.(2019); Schmitz et al.(2018) and in multivariate Corresponding author: data monitoring, like anesthesia and patient monitoring Tim Ziemer, University of Bremen, Medical Image Computing Group, Sanderson et al.(2005), stock market monitoring Neuhoff Enrique-Schmidt-Str. 5, 28359 Bremen, Germany et al.(2002), data exploration and browsing Yeung(1980); Email: [email protected] Prepared using sagej.cls [Version: 2017/01/17 v1.20] 2 Journal Title XX(X) Background when altering several parameters at once, the urgency levels do not add up, but create somewhat nonlinear effects. Orthogonality is a topic that has been treated a lot in the fields of psychoacoustics and auditory display research, and will be Hermann(2002) lists the “lack of perceptual orthogonal- briefly discussed in this section, followed by previous work. ity” as one of the most important difficulties in auditory A lot of previous work either focused on the implementation displays. He agrees with Anderson and Sanderson(2009) of psychoacoustics in sonification design or on orthogonal that auditory perception is too little understood to specify dimensions in sonification. Our work integrates these two multiple orthogonal dimensions. Likewise, Brewster(2003) lines of research by leveraging psychoacoustic knowledge to lists the “lack of orthogonality” one of the problems with sonify perceptually orthogonal dimensions. nonspeech sound. (Worrall 2009, ch. 2) agrees with this observation, too, Orthogonality stating that “(. ) parameter mapping requires a working knowledge of how the parameters interact with each other Following the literature on sonification Worrall(2009); perceptually”, because these interaction may obscure data Hermann(2002) and psychoacoustics Schneider(1997, relations and even confuse the listener Worrall(2019). He 2018a) we define dimensions in a Cartesian way as having thinks that attempts to create a perceptually orthogonal both a direction and a distance, also referred to as polarity sonification space have not yet been successful, giving the and magnitude. In that sense, information like the radius, timbre space sonification approach Barrass(1997) as an which is one dimension in polar and cylindrical coordinates, example. At the same time, he expressed the need for better is only a half-dimension; it only informs about a distance, tools. not a direction Parseihian et al.(2016). It is widely accepted that auditory sensations and other psychological attributes are never perfectly orthogonal, as the dimensions may be correlated to some extent Schneider(1997). Hence, we Psychoacoustics in Sonification Design consider dimensions as orthogonal, if a magnitude change of one dimension hardly affects the magnitude of any other The need to consider psychoacoustics in sonification design dimension, often referred to as separable Garner(1974); has been expressed in numerous studies Hellier et al.(1993); Schneider(2018b); Neuhoff(2004). In that sense, they are Hermann(2002); Bovermann et al.(2011); Bly(1982); linearly independent from one another Worrall(2009), i.e., Kramer(1994); Barrass(1994); Smith(1990); Williams they barely exhibit any coupling or perceptual interactions (1994); Ferguson and Brewster(2017); Walker and Kramer (Hermann 2002, ch. 3); Anderson and Sanderson(2004, (2004); Ferguson et al.(2006); Bliss and Spain(2007); Hunt 2009). Furthermore, a dimension must be continuous, i.e., and Hermann(2011); Degara et al.(2013b). on interval scale or ratio scale rather than ordinal or nominal Ferguson and Brewster(2017) evaluate psychoacoustic scale Schneider(1997). parameters for sonification. They argue that pitch is a meaningful dimension, as human listeners have a high Orthogonality in Sonification resolution in pitch perception. They suggest the use of There are plenty examples of sonifications mapping one half- loudness fluctuation and roughness as additional dimensions. dimension to amplitude and another to frequency Neuhoff The authors of Ferguson et al.(2006) suggest mapping of et al.(2002) and it is not surprising that the authors parameters to pitch, loudness, roughness, and brightness. realized in their evaluations that these physically orthogonal Likewise, Parseihian et al.(2016) consider pitch, loudness, dimensions interact perceptually. duration/tempo and timbre as orthogonal and as the main Anderson and Sanderson(2009) tried out several perceptual attributes of sound. Arfib et al.(2002) name pitch, mapping principles for multivariate data in complex loudness, timbre aspects, like brightness, roughness, attack work domains. They do not consider psychoacoustics time, vibrato and formants, spatialization, as well as their in their parameter mapping approach, but map multiple temporal derivatives, as psychoacoustic parameters suitable variables to physical parameters, like amplitude, amplitude for understandable multidimensional sonification. modulations, fundamental frequency, cutoff-frequency, pulse The authors of Parseihian et al.(2016) managed to width, etc. They realize that participants had problems implement and validate sonification designs